CAS/CNS Technical Reports

Permanent URI for this collection

https://boston7-test.atmire.com/handle/2144/1897

Browse

Now showing 1 - 20 of 400

PointMap: A real-time memory-based learning system with on-line and post-training pruning
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2002-12) Kopco, Norbert; Carpenter, Gail
A memory-based learning system called PointMap is a simple and computationally efficient extension of Condensed Nearest Neighbor that allows the user to limit the number of exemplars stored during incremental learning. PointMap evaluates the information value of coding nodes during training, and uses this index to prune uninformative nodes either on-line or after training. These pruning methods allow the user to control both a priori code size and sensitivity to detail in the training data, as well as to determine the code size necessary for accurate performance on a given data set. Coding and pruning computations are local in space, with only the nearest coded neighbor available for comparison with the input; and in time, with only the current input available during coding. Pruning helps solve common problems of traditional memory-based learning systems: large memory requirements, their accompanying slow on-line computations, and sensitivity to noise. PointMap copes with the curse of dimensionality by considering multiple nearest neighbors during testing without increasing the complexity of the training process or the stored code. The performance of PointMap is compared to that of a group of sixteen nearest-neighbor systems on benchmark problems.
Contrast-sensitive perceptual grouping and object-based attention in the laminar circuits of primary visual cortex
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1999-03) Grossberg, Stephen; Raizada, Rajeev D.S.
Recent neurophysiological studies have shown that primary visual cortex, or Vl, does more than passively process image features using the feedforward filters suggested by Hubel and Wiesel. It also uses horizontal interactions to group features preattentively into object representations, and feedback interactions to selectively attend to these groupings. All neocortical areas, including Vl, are organized into layered circuits. We present a neural model showing how the layered circuits in areas Vl and V2 enable feedforward, horizontal, and feedback interactions to complete perceptual groupings over positions that do not receive contrastive visual inputs, even while attention can only modulate or prime positions that do not receive such inputs. Recent neurophysiological data about how grouping and attention occur and interact in Vl are simulated and explained, and testable predictions are made. These simulations show how attention can selectively propagate along an object grouping and protect it from competitive masking, and how contextual stimuli can enhance or suppress groupings in a contrast-sensitive manner.
Physical Limits to Spatial Resolution of Optical Recording: Clarifying the Spatial Structure of Cortical Hypercolumns
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Polimeni, Jonathan; Granquist-Fraser, Domhnull; Wood, Richard; Schwartz, Eric
Neurons in macaque primary visual cortex are spatially arranged by their global topographic position and in at least three overlapping local modular systems: ocular dominance columns, orientation pinwheels, and cytochrome oxidase (CO) blobs. Individual neurons in the blobs are not tuned to orientation, and populations of neurons in the pinwheel center regions show weak orientation tuning, suggesting a close relation between pinwheel centers and CO blobs. However, this hypothesis has been challenged by a series of optical recording experiments. In this report, we show that the statistical error associated with photon scatter and absorption in brain tissue combined with theblurring introduced by the optics of the imaging system has typically been in the range of 250 μm. These physical limitations cause a systematic error in the location of pinwheel centers because of the vectorial nature of these patterns, such that the apparent location of a pinwheel center measured by optical recording is never (on average) in the correct in vivo location. The systematic positional offset is about 116 μtm, which is large enough to account for the claimed mis-alignment of CO blobs and pinwheel centers. Thus, optical recording, as it has been used to date, has insufficient spatial resolution to accurately locate pinwheel centers. The earlier hypothesis that CO blobs and pinwheel centers are co-terminous remains the only one currently supported by reliable observation.
Logic and Phenomenology of Incompleteness in Illusory Figures: New Cases and Hypotheses
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-03) Pinna, Baingio; Grossberg, Stephen
Cognitive and gestalt visions theories consider incompleteness to be a necessmy and sufficient factor for inducing illusory figures. The role of incompleteness is studied herein by defining the inner logic subtended by use of the term "incompleteness", presenting new cases to clarify the phenomenology of incompleteness as a necessary and sufficient condition, and suggesting an alternative hypothesis to explain illusory figures after analyzing problems with the incompleteness hypothesis. It is demonstrated that incompleteness is not a sufficient condition, illusory figures do not necessarily complete incompletenesses, the shape of incompleteness does not predict the shape of illusory figures, and incompleteness is not a necessary condition. Finally, it is noted that the incompleteness hypothesis can be replaced by concepts concerning interacting boundary grouping and surface filling-in processes during figure-ground segregation. The suggested hypothesis is consistent with neurophysiological experiments and is described in terms of the FACADE neural model of boundary and surface formation during figure-ground segregation.
EyeRIS: A General-Purpose System for Eye Movement Contingent Display Control
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-09) Santini, Fabrizio; Redner, Gabriel; Lovin, Ramon; Rucci, Michele
In experimental studies of visual performance, the need often emerges to modify the stimulus according to the eye movements perfonncd by the subject. The methodology of Eye Movement-Contingent Display (EMCD) enables accurate control of the position and motion of the stimulus on the retina. EMCD procedures have been used successfully in many areas of vision science, including studies of visual attention, eye movements, and physiological characterization of neuronal response properties. Unfortunately, the difficulty of real-time programming and the unavailability of flexible and economical systems that can be easily adapted to the diversity of experimental needs and laboratory setups have prevented the widespread use of EMCD control. This paper describes EyeRIS, a general-purpose system for performing EMCD experiments on a Windows computer. Based on a digital signal processor with analog and digital interfaces, this integrated hardware and software system is responsible for sampling and processing oculomotor signals and subject responses and modifying the stimulus displayed on a CRT according to the gaze-contingent procedure specified by the experimenter. EyeRIS is designed to update the stimulus within a delay of 10 ms. To thoroughly evaluate EyeRIS' perforltlancc, this study (a) examines the response of the system in a number of EMCD procedures and computational benchmarking tests, (b) compares the accuracy of implementation of one particular EMCD procedure, retinal stabilization, to that produced by a standard tool used for this task, and (c) examines EyeRIS' performance in one of the many EMCD procedures that cannot be executed by means of any other currently available device.
ClasserScript v1.1 User's Guide
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-05) Martens, Siegfried
Brain Categorization: Learning, Attention, and Consciousness
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Grossberg, Stephen; Carpenter, Gail; Ersoy, Bilgin
How do humans and animals learn to recognize objects and events? Two classical views are that exemplars or prototypes are learned. A hybrid view is that a mixture, called rule-plus-exceptions, is learned. None of these models learn their categories. A distributed ARTMAP neural network with self-supervised learning incrementally learns categories that match human learning data on a class of thirty diagnostic experiments called the 5-4 category structure. Key predictions of ART models have received behavioral, neurophysiological, and anatomical support. The ART prediction about what goes wrong during amnesic learning has also been supported: A lesion in its orienting system causes a low vigilance parameter.
Active Estimation of Distance in a Robotic Vision System that Replicates Human Eye Movement
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Santini, Fabrizio; Rucci, Michele
Many visual cues, both binocular and monocular, provide 3D information. When an agent moves with respect to a scene, an important cue is the different motion of objects located at various distances. While a motion parallax is evident for large translations of the agent, in most head/eye systems a small parallax occurs also during rotations of the cameras. A similar parallax is present also in the human eye. During a relocation of gaze, the shift in the retinal projection of an object depends not only on the amplitude of the movement, but also on the distance of the object with respect to the observer. This study proposes a method for estimating distance on the basis of the parallax that emerges from rotations of a camera. A pan/tilt system specifically designed to reproduce the oculomotor parallax present in the human eye was used to replicate the oculomotor strategy by which humans scan visual scenes. We show that the oculomotor parallax provides accurate estimation of distance during sequences of eye movements. In a system that actively scans a visual scene, challenging tasks such as image segmentation and figure/ground segregation greatly benefit from this cue.
The Watercolor Illusion and Neon Color Spreading: A Unified Analysis of New Cases and Neural Mechanisms
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Pinna, Baingio; Grossberg, Stephen
Coloration and figural properties of neon color spreading and the watercolor illusion are studied using phenomenal and psychophysical observations. Coloration properties of both effects can be reduced to a common limiting condition, a nearby color transition called the "two-dots limiting case", that clarifies their perceptual similarities and dissimilarities. The results are explained by the FACADE neural model of biological vision. The model proposes how local properties of color transitions activate spatial competition among nearby perceptual boundaries, with boundaries of lower contrast edges weakened by competition more than boundaries of higher contrast edges. This asymmetry induces spreading of more color across these boundaries than conversely. The model also predicts how depth and figure-ground effects are generated in these illusions.
A Theoretical Analysis of the Influence of Fixational Instability on the Development of Thalamocortical Connectivity
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Casile, Antonino; Rucci, Michele
Under natural viewing conditions, the physiological inotability of visual fixation keeps the projection of the stimulus on the retina in constant motion. After eye opening, chronic exposure to a constantly moving retinal image might influence the experience-dependent refinement of cell response characteristics. The results of previous modeling studies have suggested a contribution of fixational instability in the Hebbian maturation of the receptive fields of V1 simple cells (Rucci, Edelman, & Wray, 2000; Rucci & Casile, 2004). This paper presents a mathematieal explanation of our previous computational results. Using quasi-linear models of LGN units and V1 simple cells, we derive analytical expressions for the second-order statistics of thalamocortical activity before and after eye opening. We show that in the presence of natural stimulation, fixational instability introduces a spatially uncorrelated signal in the retinal input, whieh strongly influences the structure of correlated activity in the model.
Self-Organizing Hierarchical Knowledge Discovery by an Artmap Information Fusion System
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 2005-01) Carpenter, Gail; Martens, Siegfried
Classifying terrain or objects may require the resolution of conflicting information from sensors working at different times, locations, and scales, and from users with different goals and situations. Current fusion methods can help resolve such inconsistencies, as when evidence variously suggests that an object is a car, a truck, or an airplane. The methods described here define a complementary approach to the information fusion problem, considering the case where sensors and sources arc both nominally inconsistent and reliable, as when evidence suggests that an object is a car, a vehicle, and man-made. Underlying relationships among classes are assumed to be unknown to the automated system or the human user. The ARTMAP self-organizing rule discovery procedure is illustrated with an image example, but is not limited to the image domain.
A Neural Network Method for Efficient Vegetation Mapping
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-12) Carpenter, Gail; Gopal, Sucharita; Macomber, Scott
This paper describes the application of a neural network method designed to improve the efficiency of map production from remote sensing data. Specifically, the ARTMAP neural network produces vegetation maps of the Sierra National Forest, in Northern California, using Landsat Thematic Mapper (TM) data. In addition to spectral values, the data set includes terrain and location information for each pixel. The maps produced by ARTMAP are of comparable accuracy to maps produced by a currently used method, which requires expert knowledge of the area as well as extensive manual editing. In fact, once field observations of vegetation classes had been collected for selected sites, ARTMAP took only a few hours to accomplish a mapping task that had previously taken many months. The ARTMAP network features fast on-line learning, so the system can be updated incrementally when new field observations arrive, without the need for retraining on the entire data set. In addition to maps that identify lifeform and Calveg species, ARTMAP produces confidence maps, which indicate where errors are most likely to occur and which can, therefore, be used to guide map editing.
A Neural Model for Self Organizing Feature Detectors and Classifiers in a Network Hierarchy
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-11) Williamson, James R.
Many models of early cortical processing have shown how local learning rules can produce efficient, sparse-distributed codes in which nodes have responses that are statistically independent and low probability. However, it is not known how to develop a useful hierarchical representation, containing sparse-distributed codes at each level of the hierarchy, that incorporates predictive feedback from the environment. We take a step in that direction by proposing a biologically plausible neural network model that develops receptive fields, and learns to make class predictions, with or without the help of environmental feedback. The model is a new type of predictive adaptive resonance theory network called Receptive Field ARTMAP, or RAM. RAM self organizes internal category nodes that are tuned to activity distributions in topographic input maps. Each receptive field is composed of multiple weight fields that are adapted via local, on-line learning, to form smooth receptive ftelds that reflect; the statistics of the activity distributions in the input maps. When RAM generates incorrect predictions, its vigilance is raised, amplifying subtractive inhibition and sharpening receptive fields until the error is corrected. Evaluation on several classification benchmarks shows that RAM outperforms a related (but neurally implausible) model called Gaussian ARTMAP, as well as several standard neural network and statistical classifters. A topographic version of RAM is proposed, which is capable of self organizing hierarchical representations. Topographic RAM is a model for receptive field development at any level of the cortical hierarchy, and provides explanations for a variety of perceptual learning data.
Adaptive Resonance Theory
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-09) Carpenter, Gail; Grossberg, Stephen
Photo-Realistic Scenes with Cast Shadows Show No Above/Below Search Asymmetries for Illumination Direction
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-12) Cunningham, Robert K.; Beck, Jacob; Mingolla, Ennio
Visual search is extended from the domain of polygonal figures presented on a uniform field to photo-realistic scenes containing target objects in dense, naturalistic backgrounds. The target in a trial is a computer-rendered rock protruding in depth from a "wall" of rocks of roughly similar size but different shapes. Subjects responded "present" when one rock appeared closer than the rest, owing to occlusions or cast shadows, and "absent" when all rocks appeared to be at the same depth. Results showed that cast shadows can significantly decrease reaction times compared to scenes with no cast shadows, in which the target was revealed only by occlusions of rocks behind it. A control experiment showed that cast shadows can be utilized even for displays involving rocks of several achromatic surface colors (dark through light), in which the shadow cast by the target rock was not the darkest region in the scene. Finally, in contrast with reports of experiments by others involving polygonal figures, we found no evidence for an effect of illumination direction (above vs. below) on search times.
Neural Sensor Fusion for Spatial Visualization on a Mobile Robot
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-08) Martens, Siegfried; Carpenter, Gail A.; Gaudiano, Paolo
An ARTMAP neural network is used to integrate visual information and ultrasonic sensory information on a B 14 mobile robot. Training samples for the neural network are acquired without human intervention. Sensory snapshots are retrospectively associated with the distance to the wall, provided by on~ board odomctry as the robot travels in a straight line. The goal is to produce a more accurate measure of distance than is provided by the raw sensors. The neural network effectively combines sensory sources both within and between modalities. The improved distance percept is used to produce occupancy grid visualizations of the robot's environment. The maps produced point to specific problems of raw sensory information processing and demonstrate the benefits of using a neural network system for sensor fusion.
Neural Models of Normal and Abnormal Behavior: What Do Schizophrenia, Parkinsonism, Attention Deficit Disorder, and Depression Have in Common?
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-10) Grossberg, Stephen
Neural Dynamics of 3-D Surface Perception: Figure-Ground Separation and Lightness Perception
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-07) Kelly, Frank; Grossberg, Stephen
This article develops the FACADE theory of three-dimensional (3-D) vision to simulate data concerning how two-dimensional (2-D) pictures give rise to 3-D percepts of occluded and occluding surfaces. The theory suggests how geometrical and contrastive properties of an image can either cooperate or compete when forming the boundary and surface representations that subserve conscious visual percepts. Spatially long-range cooperation and short-range competition work together to separate boundaries of occluding ligures from their occluded neighbors, thereby providing sensitivity to T-junctions without the need to assume that T-junction "detectors" exist. Both boundary and surface representations of occluded objects may be amodaly completed, while the surface representations of unoccluded objects become visible through modal processes. Computer simulations include Bregman-Kanizsa figure-ground separation, Kanizsa stratification, and various lightness percepts, including the Munker-White, Benary cross, and checkerboard percepts.
Building Adaptive Basis Functions with a Continuous SOM
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-06) Campos, Marcos M.; Carpenter, Gail A.
This paper introduces CSOM, a distributed version of the Self-Organizing Map network capable of generating maps similar to those created with the original algorithm. Due to the continuous nature of the mapping, CSOM outperforms the traditional SOM algorithm in function approximation tasks. System performance is illustrated with three examples.
A Neural Model of Saccadic Eye Movement Control Explains Task-Specific Adaptation
(Boston University Center for Adaptive Systems and Department of Cognitive and Neural Systems, 1998-07) Gancarz, Gregory; Grossberg, Stephen
Multiple brain learning sites are needed to calibrate the accuracy of saccadic eye movements. This is true because saccadcs can be made reactively to visual cues, attentively to visual or auditory cues, or planned in response to memory cues using visual, parietal, and prefrontal cortex, as well as superior colliculus, cerebellum, and reticular formation. The organization of these sites can be probed by displacing a visual target during a saccade. The resulting adaptation typically shows incomplete and asymmetric transfer between different tasks. A neural model of saccadic system learning is developed to explain these data, as well as data about saccadic coordinate changes.

Browse

Recent Submissions